Cost effectiveness of aggressive care for patients with nontraumatic ...

Cost effectiveness of aggressive care for patients with nontraumatic coma* Mary Beth Hamel, MD, MPH; Russell Phillips, MD; Joan Teno, MD, MS; Roger B. Davis, ScD; Lee Goldman, MD, MPH; Joanne Lynn, MD; Norman Desbiens, MD; Alfred F. Connors Jr, MD; Joel Tsevat, MD, MPH

Objective: To estimate the cost effectiveness of aggressive care for patients with nontraumatic coma. Design: Cost-effectiveness analysis. Setting: Five academic medical centers. Patients: Patients with nontraumatic coma enrolled in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT). Patients with reversible metabolic causes of coma such as diabetic ketoacidosis or uremia were excluded. Measurements: We calculated the incremental cost effectiveness of continuing aggressive care vs. withholding cardiopulmonary resuscitation and ventilatory support after day 3 of coma. We estimated life expectancy based on up to 4.6 yrs of follow-up. Utilities (quality-of-life weights) were estimated using timetradeoff questions. Costs were based on hospital fiscal data and Medicare data. Separate analyses were conducted for two prognostic groups based on five risk factors assessed on day 3 of coma: age >70 yrs, abnormal brainstem response, absent verbal response, absent withdrawal to pain, and serum creatinine >132.6 ␮mol/L (1.5 mg/dL).

O

Results: For the 596 patients studied, the median (25th, 75th percentile) age was 67 yrs (range, 55–77), and 52% were female. By 2 months after enrollment, 69% had died, 19% were severely disabled, 7% had survived without severe disability, and 4% had survived with unknown functional status. The incremental cost effectiveness of the more aggressive care strategy was $140,000 (1998 dollars) per quality-adjusted life year (QALY) for high-risk patients (3–5 risk factors, 93% 2-month mortality) and $87,000/ QALY for low-risk patients (0 –2 risk factors, 49% mortality). In sensitivity analyses, the incremental cost per QALY did not fall below $50,000/QALY, even with wide variation in our baseline estimates. Conclusions: Continuing aggressive care after day 3 of nontraumatic coma is associated with a high cost per QALY gained, especially for patients at high risk for poor outcomes. Earlier decisions to withhold life-sustaining treatments for patients with very poor prognoses may yield considerable cost savings. (Crit Care Med 2002; 30:1191–1196) KEY WORDS: coma; cost effectiveness; medical decision making; outcomes; costs cardiopulmonary resuscitation

utcomes for patients with nontraumatic coma are generally poor; the majority of patients die within 1 month, and a substantial proportion of survivors are left with severe functional impairments (1–3). Families and physicians often face difficult decisions about whether and when to limit expensive and invasive life-sustaining treatments.

The cost effectiveness of aggressive care for nontraumatic coma has not been studied. We anticipated that the incremental cost effectiveness of aggressive treatment strategies for nontraumatic coma would vary according to patients’ baseline prognoses. Previously, we had identified risk factors associated with poor outcomes and developed a prognostic system to identify

*See also p. 1382. From the Division of General Medicine and Primary Care (MBH, RP, RBD), Department of Medicine, Beth Israel Deaconess Medical Center, Boston, MA; Brown University (JoaT), Providence, RI; Department of Medicine (LG), University of California, San Francisco, School of Medicine; the Center to Improve Care of the Dying (JL), RAND, DC; Chattanooga Unit of the University of Tennessee College of Medicine (ND), Chattanooga, TN; University of Virginia School of Medicine (AFC), Charlottesville, VA; and the Section of Outcomes Research (JoeT), Division of General Internal Medicine, Department of Internal Medicine, and the Center for Clinical Effectiveness, Institute for Health Policy and

Health Services Research, University of Cincinnati Medical Center, Cincinnati, OH. Supported by the Robert Wood Johnson Foundation. The opinions and findings contained in this manuscript are those of the authors and do not necessarily represent the views of the Robert Wood Johnson Foundation or their Board of Trustees. Dr. Hamel is supported by a Career Development Award from the National Institute on Aging (K08 AG0075-02). Presented, in part, at the National Meeting of the Society of General Internal Medicine, San Francisco, CA, April 1999. Copyright © 2002 by Lippincott Williams & Wilkins

Crit Care Med 2002 Vol. 30, No. 6

patients who are unlikely to survive (1). For our present analysis, we studied a cohort of patients hospitalized with nontraumatic coma (excluding more reversible causes of coma such as diabetic ketoacidosis and uremia) and used our prognostic model to divide patients according to their risk of short-term mortality. Within each prognostic group, we divided patients based on whether decisions had been made by day 4 of coma to withhold cardiopulmonary resuscitation (CPR) and ventilatory support (i.e., whether patients had do-not-resuscitate [DNR] and do-not-intubate orders) and estimated the incremental cost effectiveness of continuing aggressive treatment vs. withholding CPR and ventilatory support.

METHODS Overview We compared the costs and outcomes of two treatment strategies for patients with

1191

nontraumatic coma: a) a less aggressive treatment strategy involving a decision by day 4 of coma to withhold CPR and ventilatory support (i.e., DNR orders had been written) and b) A more aggressive treatment strategy, which did not include a plan to limit these life-sustaining treatments. We stratified patients according to an objective assessment of their prognoses and estimated the incremental cost effectiveness of the more aggressive treatment strategy compared with the less aggressive treatment strategy.

Patient Population This cost-effectiveness analysis is based on data from patients with nontraumatic coma enrolled in the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT), a prospective study of outcomes, preferences, and decision making for seriously ill hospitalized adults. Full descriptions of the objectives and methods of SUPPORT have been published previously (4, 5). Briefly, inpatients were enrolled prospectively from June 1989 through January 1994 at five academic medical centers. Institutional review boards of the participating hospitals approved the study design, and informed consent was obtained orally before interviews with patients, their families, and their physicians. SUPPORT had two phases: an observational phase and an interventional phase. During the interventional phase, clinicians randomized to the intervention were given information about patients’ prognoses and preferences for care, and nurses facilitated communication between patients and clinicians. Because the intervention did not affect mortality or resource use (5), patients from both phases and from both the intervention and control groups were included in our analysis. Patients were eligible for SUPPORT if they were ⱖ18 yrs and met criteria for at least one of nine diagnostic categories (4), one of which was nontraumatic coma. Patients who were discharged from the hospital or who died within 48 hrs of study entry were excluded from SUPPORT. Patients were included in the coma category if during the first 24 hrs of their hospitalization or while receiving care in the intensive care unit they were described by a physician as “comatose,” “unconscious,” or “obtunded” and had a Glasgow Coma Score ⱕ9 for ⱖ6 hrs (2, 6). Patients were excluded if trauma, drug intoxication, hypothermia, or an operative complication was the primary cause of coma. Patients were also excluded if they had any of the following metabolic causes of coma: diabetic ketotacidosis, nonketotic hyperosmolar syndrome, thyrotoxicosis, myxedema, hepatic encephalopathy, uremia, hyponatremia, hypernatremia, hypocalcemia, or

1192

hypercalcemia. Patients with coma due to hypoglycemia or hypoxemia were included.

Data Collection Research nurses reviewed medical records throughout patients’ hospitalizations and abstracted patients’ diagnoses and comorbid conditions. They also recorded data for calculating the Therapeutic Intervention Scoring System score (7) on study days 1, 3, 7, 14, and 25. The Therapeutic Intervention Scoring System is a measure of resource intensity that assigns one point for minor interventions (such as pulse oximetry or peripheral intravenous therapy) and two to four points for more substantial interventions (such as intubation or surgery). Comorbid conditions (such as diabetes mellitus, congestive heart failure, and stroke) were abstracted using a list developed as part of the Acute Physiology and Chronic Health Evaluation II scoring system; a comorbidity score was calculated by a simple count of comorbid conditions (8). Nurses reviewed hospital charts to determine whether decisions were made to withhold CPR or ventilatory support. Trained interviewers interviewed patients and their surrogates to obtain information on patients’ functional status (as measured by a modified version of the Katz Activities of Daily Living Scale) (9, 10) and utilities (see below). The surrogate was defined as the person who would make decisions on the patient’s behalf if the patient were unable to do so. Patients were excluded from interviews if they were unable to communicate because of coma, having an endotracheal tube, cognitive impairment, or other reasons.

Analytic Methods Descriptive Statistics and Prognostic Model. To characterize patients at baseline and at the 2-month follow-up, we tabulated frequencies for categorical variables and calculated median and 25th and 75th percentiles for continuous variables. Statistical analyses were preformed using SAS (SAS Institute, Cary, NC). Using a prognostic model we previously developed and validated (1), we stratified patients into two groups based on their risk of dying within 2 months. The prognostic model assigns one point for each of the following five risk factors on day 3 of coma: age ⱖ70 yrs, abnormal brainstem response, absent verbal response, absent withdrawal to pain, and serum creatinine level ⱖ132.6 ␮mol/L (1.5 mg/dL). Based on the 222 patients used to derive the prognostic model, predictions of 2-month mortality according to number of risk factors were as follows: zero risk factors, 20%; one risk factor, 33%; two risk factors, 60%; three risk factors, 97%; four risk factors, 96%; and five risk factors, 100%. For the purpose of this cost-effectiveness anal-

ysis, we defined the high-risk group as having three to five risk factors and the low-risk group as having zero to two risk factors. To determine whether differences in patients’ prognoses within the two prognostic strata (0 –2 vs. 3–5 risk factors) guided treatment decisions, we compared the mean number of risk factors for patients who were treated aggressively with those for whom decisions were made by study day 4 to withhold CPR and ventilatory support. To examine whether our definition of aggressive care captured differences in intensity of treatment, we compared day 7 resource intensity (Therapeutic Intervention Scoring System) scores according to whether patients had DNR and donot-intubate orders. Utilities. In cost-effectiveness analyses, utilities (quality-of-life weights) are used to adjust quantity of life for its quality and to allow years spent in health states associated with worse quality of life to be valued less than health states associated with better quality of life. Utility scores have values from zero to one, with higher scores indicating better quality of life. The years of life gained by a particular medical intervention are multiplied by the utility associated with those years to estimate the quality-adjusted life years (QALYs) gained by that medical intervention. We used patients’ and surrogates’ responses to time-tradeoff questions asked during interviews at the 6-month follow-up to estimate their utilities (11–13). For the timetradeoff questions, interviewers asked patients to state the amount of time in excellent health they would equate with living 12 months in their current state of health. For example, a patient willing to trade 12 months in his current health for no less than 6 months in perfect health would have a utility of 0.5 (6/12). For our baseline analysis, we assumed that the utilities reported by interviewed patients and surrogates at the 6-month follow-up represented those of all patients for their remaining months or years, and we used the overall mean value to quality-adjust life expectancy for both risk groups. Healthcare Costs. Charges for the index hospitalizations were collected from the participating hospitals’ billing systems. We estimated hospital costs by adjusting charges using Medicare cost-to-charge ratios for Uniform Bill 1982 (UB82) cost centers at each participating hospital. We focused on the cost of the index hospitalization incurred after study day 3 because our analyses compared costs and outcomes experienced after the decisions made by study day 4 concerning whether to withhold resuscitation and ventilatory support. We used primary data to estimate costs separately according to risk group and treatment strategy. We linked Medicare financial data (Parts A and B) with SUPPORT data to estimate hospitalization costs after the index hospitalization and to estimate physician costs during and after the index hospitalization. We assumed


that costs for patients without Medicare insurance (i.e., most patients ⬍65 yrs) were similar to those of patients covered by Medicare. We estimated physician costs using ResourceBased Relative Value Scale methodology (14). We estimated annual hospital and physician costs after year 1 based on costs incurred during the final quarter of year 1. We assumed that patients with four or more dependencies in activities of daily living were severely disabled (1) and would require long-term care. (For high-risk patients, 94% had four or more dependencies in the Activities of Daily Living Scale, and 70% of low-risk patients had four or more dependencies; therefore, we assumed that these percentages of survivors would require long-term care.) We estimated the annual cost of long-term care based on the Tung et al. (15) estimate of the costs of institutional care for patients who were severely disabled after suffering a stroke. Their estimate was within $300 dollars of an estimate made by Leon and colleagues (16) of the annual costs of caring for a person with severe Alzheimer’s disease. We converted all estimated costs into 1998 dollars using the medical component of the Consumer Price Index. Our analyses only considered healthcare costs and did not attempt to estimate additional societal costs or savings. For physician and hospital costs for the index hospitalization, we used primary data to make separate estimates based on risk group and treatment strategy. For hospital and physician costs from discharge after the index hospitalization through 1 yr, we made separate estimates for each risk group, but assumed that costs would be the same for each treatment strategy. For hospital, physician and institutional care costs after year 1, we used one estimate for all patients and assumed that costs would not differ by risk group or treatment strategy. Calculating Costs per Quality-Adjusted Life-Year. Cost-effectiveness analysis is a tool that estimates the healthcare costs and health benefits associated with various medical interventions. The benefits are measured in gains in quantity and quality of life and are expressed as QALYs. Cost-effectiveness ratios (expressed as dollars per QALY gained) represent the additional costs and additional quality-adjusted life expectancy associated with one medical treatment compared with a less effective treatment. We calculated the following ratio to evaluate the cost effectiveness of aggressive care vs. withholding CPR and ventilatory support: cost of more aggressive care ⫺ cost of less aggressive care $ ⫽ , QALY QALE of more aggressive care ⫺ QALE of less aggressive care where QALE is quality-adjusted life expectancy. We conducted our cost-effectiveness analyses from a societal perspective. Cost-effective-


ness analyses were done using DATA 3.0 (TreeAge Software, Williamstown, MA). We used Markov models (17) to estimate expected lifetime costs and quality-adjusted life expectancy for each strategy (i.e., aggressive care vs. withholding CPR and ventilatory support). A Markov model tracks a cohort over time until the entire cohort has died. The model requires specification of the probability of dying each year (i.e., 1-yr mortality estimates). We estimated life expectancy based on complete follow-up of SUPPORT patients through December 31, 1994, using the National Death Index. Our mortality estimates for years 1, 2, and 3 were derived directly from Kaplan-Meier curves for SUPPORT patients. Mortality estimates for year 1 were made for each risk group, broken down by treatment strategy. Annual mortality after year 1 was based on a survival curve for all patients because there was not a sufficient number of survivors to make separate estimates according to risk group or treatment strategy. Therefore, we made the simplifying assumption that survival after year 1 was not related to risk group or initial treatment strategy. We estimated mortality after year 3 by summing disease-specific mortality (i.e., excess mortality associated with having a history of nontraumatic coma) and age-specific mortality (i.e., the background mortality expected if patients did not have nontraumatic coma). Age-specific mortality was based on U.S. life tables (18). Note that age-specific mortality increases each year as the cohort ages. We estimated diseasespecific mortality from mortality observed in year 3 of SUPPORT. (Data after year 3 were not used due to flattening of the survival curves and small sample size after year 3.) We isolated disease-specific mortality from total mortality by subtracting out age-specific mortality in year 3. We assumed that disease-specific mortality would remain constant in subsequent years. To estimate total mortality for each year after year 3, we added disease-specific mortality to age-specific mortality. To estimate the total cost of each treatment strategy for each prognostic group, we summed the mean hospital and physician costs for the index hospitalization, mean hospital and physician costs from discharge through 1 yr of follow-up, mean annual hospital and physician costs after year 1 (based on Part A and B Medicare costs for quarter 4 of year 1), and mean annual cost of long-term care. We discounted all costs and qualityadjusted years at a rate of 3% per year (19). We calculated the incremental cost effectiveness in 1998 dollars per quality-adjusted life-year (QALY) gained of the more aggressive vs. the less aggressive treatment strategy for each of the two risk groups. We rounded incremental

cost-effectiveness ratios to two significant figures.

Sensitivity Analyses We performed sensitivity analyses to assess how different assumptions would affect our results. In one-way sensitivity analyses, we individually varied each of our cost estimates and our estimate for annual mortality after year 1 from 50% to 200% of our baseline estimates. In addition, we varied the utilities from 0.5 to 1.0 and the discount rate from 0% to 10%. We also performed modified best- and worst-case three-way sensitivity analyses, where we simultaneously varied the three variables to which our results were most sensitive (annual mortality after year 1, year 1 healthcare costs, and annual costs after year 1). To bias our analysis toward finding more favorable cost-effectiveness ratios, we simultaneously lowered these three estimates to 67% of our baseline values, and to bias our analysis toward finding less favorable ratios, we simultaneously increased these three estimates to 150% of our baseline values.

RESULTS Patients’ Baseline Characteristics and Outcomes. Within 2 months after enrollment, 413 of the 596 patients with nontraumatic coma (69%) had died, 115 (19%) were severely disabled (4 –7 dependencies in Activities of Daily Living Scale), six (1%) had survived with mildmoderate disability (2–3 dependencies), 37 (6%) had survived with no or mild disability (0 –1 dependencies), and 25 (4%) had survived with unknown functional status (Table 1). Information about the five risk factors included in the prognostic model was available for 549 patients: 290 (53%) had zero to two risk factors and were classified as low-risk, and 259 (47%) had three to five risk factors and were classified as high-risk. By study day 4, decisions were made to withhold CPR and ventilatory support (i.e., the less aggressive care strategy was undertaken) for 40 (14%) of the 290 lowrisk patients and for 81 (31%) of the 259 high-risk patients. For high-risk patients, the mean number of risk factors was similar for patients who were treated aggressively vs. those not treated aggressively, suggesting that DNR orders were not a marker for poorer prognosis (mean number of risk factors 3.6 for patients treated aggressively and 3.5 for patients who had DNR orders). For low-risk patients, those who had a DNR order had modestly worse prognoses, with a mean number of risk 1193

factors of 1.6 compared with 1.3 for those in the aggressive care strategy (p ⫽ .005). The two care strategies we defined based on decisions to withhold CPR and ventilatory support described groups of patients who received different intensities of treatment. For low-risk patients, the mean day 7 resource intensity (Therapeutic Intervention Scoring System) score was 12.0 for those who were DNR/do-notintubate by day 4 and 27.7 for those who were not (p ⫽ .0001). For high-risk patients, the mean resource intensity score was 14.9 compared with 26.2 (p ⫽ .003). First-year costs were considerably higher for patients who were treated aggressively (Table 2). Two-month mortality was 49% for low-risk patients and 93% for high-risk patients (Fig. 1). For low-risk patients, 1-yr mortality was 58% for patients receiving more aggressive care and 95% for patients receiving less aggressive care. For high-risk patients, 1-yr mortality was 95% for patients who received more aggressive care and 99% for patients receiving less aggressive care. Based on all patients who survived at least 1 yr (not divided according to treatment strategies or risk group), the annual disease-related mortality after year 1 was estimated to be 7.5%. Of patients who survived at least 6 months, 31 patients and 70 surrogates provided data on patient utilities. The mean utility for this group was 0.68. Cost-Effectiveness Analyses: Baseline Case. The incremental cost per QALY gained of providing aggressive care rather than withholding CPR and ventilatory support by day 4 of coma was $140,000 (1998 dollars) for high-risk patients and $87,000 for low-risk patients. Sensitivity Analyses. In one-way sensitivity analyses, even with wide variations of our baseline estimates, incremental cost-effectiveness ratios did not fall below $50,000/QALY (Table 3). In our best-case analysis where we simultaneously lowered our estimates for annual mortality after year 1, year 1 healthcare costs, and annual costs after year 1 to 67% of our baseline assumptions, costeffectiveness ratios dropped to $57,000 for low-risk patients and $91,000 for high-risk patients. In our worst-case analysis where we simultaneously increased our estimates for these three variables to 150% of our baseline assumptions, cost-effectiveness ratios rose to $130,000 for low-risk patients and $230,000 for high-risk patients. 1194

DISCUSSION We estimated the cost effectiveness of aggressive treatment of patients with nontraumatic coma, as opposed to withholding CPR and ventilatory support by day 4 of coma. The cost effectiveness of the aggressive strategy differed according to patients’ baseline prognoses. For patients in the high-risk group (patients with three or more of the five risk factors included in a validated prognostic model), the incremental cost per QALY of the more aggressive treatment strategy was $140,000 (1998 dollars). For patients with better prognoses (0 –2 risk factors), the incremental cost was $87,000/QALY. These ratios are considerably higher than cost-effectiveness ratios for many other commonly used medical interventions. For example, coronary artery bypass surgery rather than medical therapy for left main coronary artery disease costs about $9,500/QALY (1998 dollars) (20, 21), medical therapy for severe hypertension costs about $28,000/QALY (1998 dollars) (21, 22), and treatment with t-PA rather

than streptokinase for myocardial infarction costs about $39,000 per year of life saved (1998 dollars) (23). Previous cost-effectiveness analyses have demonstrated that the cost effectiveness of aggressive treatment strategies for patients hospitalized with serious illness is dependent on patients’ baseline prognoses. In analyses using similar methods and also based on data from SUPPORT, we determined the cost effectiveness of aggressive treatment for patients with acute respiratory failure who require ventilatory support and for patients with renal failure requiring dialysis. For patients with acute respiratory failure at highest risk for short-term mortality (⬍50% probability of surviving at least 2 months), ventilator support and intensive care costs more than $100,000 per additional QALY gained. For patients who fall into the low- and medium-risk groups, the incremental cost per QALY is $29,000 and $44,000, respectively (24). When comparing initiating dialysis and continuing aggressive care with with-

Table 1. Baseline characteristics and 2-month outcomes of patients with nontraumatic coma (n ⫽ 596) Age, median (25th, 75th percentile), yr Women, No. (%) Cause of coma, No. (%) Cardiac arrest Intracerebral hemorrhage Cerebral infarction Subarachnoid hemorrhage Sepsis Central nervous system infection Mass lesions Hypoglycemia Hypovolemic shock Other/unknown cause Number of comorbid illnesses, median (25th, 75th percentile) 2-Month outcomes, No. (%) Died Survived with severe disabilitya Survived without severe disability Survived with unknown functional status a

67 (55, 77) 311 (52) 184 (31) 128 (21) 105 (18) 68 (11) 56 (9) 34 (6) 16 (3) 7 (1) 5 (1) 87 (15) 1 (0, 2)

413 (69) 115 (19) 43 (7) 25 (4)

Severe disability is defined as dependency in four or more activities of daily living.

Table 2. Baseline cost estimates (1998 dollars) Low-Risk High-Risk Patients Patients Mean hospital cost for index hospitalization (after study day 3) Aggressive care strategy Withholding aggressive care strategy Mean physician costs for index hospitalization Aggressive care strategy Withholding aggressive care strategy Mean hospitalization and physician costs from discharge through year 1 Mean annual costs after year 1 Mean annual costs for institutional care for patients, given severe disability

$38,039 $ 7,610

$16,251 $ 3,202

$ 5,159 $ 1,694 $21,723 $24,725 $38,649

$ 3,058 $ 1,287 $ 5,866 $24,725 $38,649


holding dialysis and providing comfort care, the incremental cost effectiveness of the aggressive treatment strategy is $310,000 for patients in the worst prognostic group (ⱕ10% probability of surviving at least 6 months) but $71,000 for patients in the best prognostic group (⬎40% probability of surviving at least 6 months) (25). Prior studies have examined the costs and outcomes associated with severe brain injury and the costs of life-sustaining treatments such as CPR. In a study of

patients with traumatic coma, charges per favorable functional outcome (defined by the Glasgow Outcome Scale) were $1,540,971 for patients ⱖ60 yrs and $154,155 for patients aged 20 to 40 yrs (26). In a study of 515 patients with coma admitted to a French hospital between 1987 and 1993, the treatment costs varied widely and were associated with patient age, cause of coma, and severity of illness (27). Researchers have evaluated cost savings associated with relatively early DNR

QALY, quality-adjusted life year.

orders and have documented the high costs associated with aggressive care, such as CPR. In a retrospective study of 852 patients who died in the hospital, published in 1993, average combined physician and hospital charges were $10,631 for patients who had a DNR order at admission and $73,055 for those who had a DNR order written during the hospitalization (28). In a review of the literature published in 1996, Lee et al. (29) reported that the patient’s underlying condition affected the cost effectiveness of CPR and estimated that the average cost of CPR programs was $406,605 per life saved and $225,892 per QALY saved. A study published in 1994 modeled the cost effectiveness of inpatient CPR and demonstrated that the cost per survivor increased exponentially as the rate of survival to discharge decreased ($117,000 for a rate of survival to hospital discharge of 10%, $248,271 for a rate of 1%, and $544,521 for a rate of 0.2%). Sensitivity analyses showed that stratifying hospitalized patients according to their prognosis and restricting use of CPR to those with better prognoses would substantially reduce healthcare costs (30). Our study has several limitations. We focused on decision making by day 4 of coma because accurate prognostic estimates can be made using clinical information readily available by day 3. Whereas we focused on common and clinically relevant decisions faced by families and physicians, our cost-effectiveness analysis may oversimplify the complexity of real-life medical decisions. Decisions to use or withhold a wide variety of life-sustaining treatments occur throughout patients’ illnesses. We did not attempt to control for care decisions made later in patients’ hospitalizations; the data we used reflect the actual experiences of the patients in our study. Although a minority of the patients we studied had DNR orders written by day 4 of coma, the vast majority (86%) of patients who died within 2 months had DNR orders written before death. Our results, therefore, reflect a comparison between making earlier decisions to withhold life-sustaining treatments and not making or delaying such decisions. We estimated healthcare costs incurred after the index hospitalization using Medicare data and assumed that costs for patients without Medicare insurance (i.e., most patients ⬍65 yrs) were similar. In previous SUPPORT analyses, older age


1195

Figure 1. Survival by risk group for low-risk patients (0 –2 risk factors; solid line) and high-risk patients (3–5 risk factors; dashed line).

Table 3. Baseline results and results of sensitivity analyses Incremental Cost Effectiveness (1998 Dollars/QALY)

Baseline case Annual mortality after year 1 50% of baseline estimate 200% of baseline estimate Year 1 health care costs 50% of baseline estimate 200% of baseline estimate Annual costs after year 1 50% of baseline estimate 200% of baseline estimate Utility 1 0.5 Discount rate 0% 10%

Low-Risk Patients

High-Risk Patients

$ 87,000

$140,000

$ 85,000 $ 92,000

$130,000 $160,000

$ 77,000 $110,000

$110,000 $250,000

$ 54,000 $150,000

$110,000 $210,000

$ 60,000 $120,000

$ 96,000 $190,000

$ 85,000 $ 92,000

$130,000 $160,000

T

hese limitations not withstanding, we found that con-

tinuing aggressive care after day 3 of nontraumatic coma (rather than restricting the use of life-sustaining treatments) for patients who have multiple risk factors for poor outcomes is associated with a very high incremental cost per quality-adjusted life year.

was (paradoxically) associated with lower costs (31); therefore, we may have slightly underestimated the costs for the full cohort, whose median age was 67 yrs. We had to project hospital costs after year 1 from cost data for the fourth quarter of year 1, but sensitivity analyses demonstrated that our conclusions were not sensitive to wide variations in estimates of these downstream costs. Patients were excluded from our study if they had coma caused by reversible metabolic abnormalities, and, therefore, our results are not generalizable to all patients with nontraumatic coma. Our estimates of survival, quality of life, and initial hospital costs were based on data from five medical centers; clinical outcomes and healthcare costs for patients treated at other settings, such as community hospitals, may differ. Finally, to the extent that medical care is withdrawn on the basis of poor prognosis, prognostic models can have a self-fulfilling prophecy effect. These limitations not withstanding, we found that continuing aggressive care after day 3 of nontraumatic coma (rather than restricting the use of life-sustaining treatments) for patients who have multiple risk factors for poor outcomes is associated with a very high incremental cost per QALY. Our results suggest that making decisions in the first few days of coma to withhold life-sustaining treat-

1196

ments for patients with very poor prognoses may yield considerable cost savings.

REFERENCES 1. Hamel MB, Goldman L, Teno J, et al: Identification of comatose patients at high risk for death or severe disability. SUPPORT Investigators: Understand Prognoses and Preferences for Outcomes and Risks of Treatments. JAMA 1995; 273:1842–1848 2. Levy DE, Bates D, Caronna JJ, et al: Prognosis in nontraumatic coma. Ann Intern Med 1981; 94:293–301 3. Longstreth WT Jr, Inui TS, Cobb LA, et al: Neurologic recovery after out-of-hospital cardiac arrest. Ann Intern Med 1983; 98: 588 –592 4. Lynn J, Knaus WA. Background for SUPPORT: J Clin Epidemiol 1990; 43 (Suppl): 1S– 4S 5. The SUPPORT Principal Investigators: A controlled trial to improve care for seriously ill hospitalized patients: The study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT). JAMA 1995; 274:1591–1598 6. Teasdale G, Jennett B: Assessment of coma and impaired consciousness: A practical scale. Lancet 1974; 2:81– 84 7. Cullen DJ: Therapeutic intervention scoring system: A method for quantitative comparison of patient care. Crit Care Med 1985; 13:818 – 829 8. Knaus WA, Draper EA, Wagner DP, et al: APACHE II: A severity of disease classification system. Crit Care Med 1985; 13: 818 – 829 9. Katz S, Akpom CA: A measure of primary sociobiological functions. Int J Health Serv 1976; 6:493–508 10. Katz S, Ford AB, Moskowitz RW, et al: Studies of illness in the aged: The index of ADL: A standardized measure of biological and pyschosocial function. JAMA 1963; 185: 914 –919 11. Torrance GW, Thomas WH, Sackett DL: A utility maximization model for evaluation of health care programs. Health Serv Res 1972; 7:118 –133 12. Tsevat J, Dawson NV, Matchar DB: Assessing quality of life and preferences in the seriously ill using utility theory. J Clin Epidemiol 1990; 43 (Suppl):73S–77S 13. Tsevat J, Cook EF, Green ML, et al: Health values of the seriously ill: SUPPORT investigators. Ann Intern Med 1995; 122:514 –520 14. Hsiao WC, Braun P, Becker ER, et al: The Resource-Based Relative Value Scale: Toward the development of an alternative physician payment system. JAMA 1987; 258:799 – 802 15. Tung CY, Granger CB, Sloan MA, et al: Effects of stroke on medical resource use and costs in acute myocardial infarction. Circulation 1999; 99:370 –376 16. Leon J, Cheng CK, Neuman PJ: Alzheimer’s

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

disease care: Costs and potential savings. Health Aff 1998; 17:206 –216 Sonnenberg FA, Beck JR: Markov models in medical decision making: A practical guide. Med Decis Making 1993; 13:322–338 National Center for Health Statistics: Vital Statistics of the United States, 1988: Vol. II, Sect. 6 Life Tables. Washington, DC, United States Public Health Service, 1990 Gold MR, Siegel LB, Russell LB, et al: CostEffectiveness in Health and Medicine. New York, NY, Oxford University Press, 1996 Weinstein MC, Stason WB: Cost-effectiveness of coronary artery bypass surgery. Circulation 1982; 66:III56 –III66 Goldman L: Cost-effective strategies in cardiology. In: Heart Disease: A Textbook of Cardiovascular Medicine. 4th Ed. Braunwald E (Ed). Philadelphia, PA, WB Saunders, 1992 Littenberg B, Garber AM, Sox HC Jr: Screening for hypertension. Ann Intern Med 1990; 112:192–202 Mark DB, Hlatky MA, Califf RM, et al: Cost effectiveness of thrombolytic therapy with tissue plasminogen activator as compared with streptokinase for acute myocardial infarction. N Engl J Med 1995; 332:1418 –1424 Hamel MB, Phillips RS, Davis RB, et al: Outcomes and cost-effectiveness of ventilator support and aggressive care for patients with acute respiratory failure due to pneumonia or acute respiratory distress syndrome. Am J Med 2000; 109:614 – 620 Hamel MB, Phillips RS, Davis RB, et al: Outcomes and cost-effectiveness of initiating dialysis and continuing aggressive care in seriously ill hospitalized adults. SUPPORT Investigators: Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments. Ann Intern Med 1997; 127:195–202 Pennings JL, Bachulis BL, Simons CT, et al: Survival after severe brain injury in the aged. Arch Surg 1993; 128:787–793 Sailly JC, Danze F, Lenne X, et al: Attempt to establish a classification of patients suffering from coma and admitted to a hospital structure for short or medium term treatment: Medical aspects and costs of hospital care. Acta Neurologica Belgica 1997; 97:216 –227 Maksoud A, Jahnigen DW, Skibinski CI: Do not resuscitate orders and the cost of death. Arch Intern Med 1993; 153:1249 –1253 Lee KH, Angus DC, Abramson NS: Cardiopulmonary resuscitation: What cost to cheat death? Crit Care Med 1996; 24:2046 –2052 Ebell MH, Kruse JA: A proposed model for the cost of cardiopulmonary resuscitation. Med Care 1994; 32:640 – 649 Hamel MB, Phillips RS, Teno JM, et al: Seriously ill hospitalized adults: Do we spend less on older patients? Support Investigators: Study to Understand Prognoses and Preference for Outcomes and Risks of Treatments. J Am Geriatr Soc 1996; 44:1043–1048
